[PATCH net-next 2/3] net: mdio: mdio-bcm-unimac: Poll for completion at 1us granularity

From
Florian Fainelli <florian.fainelli@broadcom.com>
Date
2026-11-13 21:16:23
Also in
lkml
Reply to
Florian Fainelli, 2026-11-13
The current polling code unconditionally waits 30us, which was sized for
a ~25us Clause 22 transaction, and then polls MDIO_CMD with a sleep of up
to 2ms between samples.

Adding Clause 45 support means that every access becomes two back to back
transactions (an ADDR cycle followed by a READ/WRITE cycle), so a fixed
delay sized for a single C22 transaction is no longer meaningful, and a
sleep of up to 2ms between samples can add significant latency to each
access when the transaction has not completed at the first sample.

Instead, poll the MDIO_START_BUSY bit every microsecond so that the
completion is noticed as soon as it happens, regardless of the type of
transaction. This turns the wait into a busy wait, which is bounded by a
timeout reduced from 100ms to 1ms: a 64 bit MDIO frame takes about 33us at
the lowest MDIO clock the driver can program from a 250MHz reference clock,
so even a two frame C45 access completes well within that.

Signed-off-by: Florian Fainelli <florian.fainelli@broadcom.com>
Co-developed-by: Carlos Soto <redacted>
Signed-off-by: Carlos Soto <redacted>
---
 drivers/net/mdio/mdio-bcm-unimac.c | 23 +++++++++++++++--------
 1 file changed, 15 insertions(+), 8 deletions(-)
diff --git a/drivers/net/mdio/mdio-bcm-unimac.c b/drivers/net/mdio/mdio-bcm-unimac.c
index 6df824aba75e..15a9880c3573 100644
--- a/drivers/net/mdio/mdio-bcm-unimac.c
+++ b/drivers/net/mdio/mdio-bcm-unimac.c
@@ -6,8 +6,8 @@
  */
 
 #include <linux/clk.h>
-#include <linux/delay.h>
 #include <linux/io.h>
+#include <linux/iopoll.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/of.h>
@@ -36,6 +36,13 @@
 #define  MDIO_CLK_DIV_MASK	0x3F
 #define  MDIO_SUPP_PREAMBLE	(1 << 12)
 
+/* A 64 bit MDIO frame takes ~25us at 2.5MHz and ~33us at the lowest MDIO
+ * clock the driver can program from a 250MHz reference clock. A C45 access is
+ * two frames, so a 1ms timeout leaves a wide margin.
+ */
+#define MDIO_POLL_DELAY_US	1
+#define MDIO_POLL_TIMEOUT_US	1000
+
 struct unimac_mdio_priv {
 	struct mii_bus		*mii_bus;
 	void __iomem		*base;
@@ -79,14 +86,14 @@ static int unimac_mdio_poll(void *wait_func_data)
 	struct unimac_mdio_priv *priv = wait_func_data;
 	u32 val;
 
-	/*
-	 * C22 transactions should take ~25 usec, will need to adjust
-	 * if C45 support is added.
+	/* C22 transactions should take ~25 usec, C45 transactions take about
+	 * twice as long since they need an address and a data cycle.
 	 */
-	udelay(30);
-
-	return read_poll_timeout(unimac_mdio_readl, val, !(val & MDIO_START_BUSY),
-				 2000, 100000, false, priv, MDIO_CMD);
+	return read_poll_timeout_atomic(unimac_mdio_readl, val,
+					!(val & MDIO_START_BUSY),
+					MDIO_POLL_DELAY_US,
+					MDIO_POLL_TIMEOUT_US, false, priv,
+					MDIO_CMD);
 }
 
 static int unimac_mdio_read(struct mii_bus *bus, int phy_id, int reg)
-- 
2.34.1
Thread (13 messages) 13 messages, 4 authors, last reply 2h ago
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help